On the Bartels Technique for Time-Series Analysis, and its Relation to the Analysis of Variance
Lawrence R. Hafstad · Journal of the American Statistical Association · 1940
IT HAS long been recognized that there is difficulty in applying the usual statistical techniques when there is correlation between successive items in any given sample. This trouble arises in many different fields: in economic time series the index number for any one day, month, or year is dependent on that of the preceding time unit. In sociology the number of strikes in progress at any one time is dependent on the number at some immediately preceding time. In agricultural experiments the yield of any plot will be high or low with the fertility of the adjacent region. This characteristic is known by various names, among statisticians as serial correlation, among physicists as nachwirkung or persistence. Various devices are used by statisticians to circumvent it, but the problem is by no means solved. For this reason it may not be amiss to bring to the attention of statisticians an approximate intuitive device used on similar problems by physicists, with the hope not only that the method in its present form might prove of some value to statisticians, but also that they may be stimulated into a more critical study of the technique.